#spontaneous hives | Explore Tumblr Posts and Blogs

unrealisticlea · 5 months

Text

collecting headcanons that spontaneously arise from the bucktommy hive mind:

Tommy calls Buck “baby”

Buck’s the one who initiates their second kiss

Tommy has a dog/cat

Tommy is Italian (?) and has a nonna who taught him how to cook pasta (I have no idea where all of this came from but I’ve read it multiple times)

however Buck is a better cook than he is

On their third date they fly together on some body of water

He does try to teach Buck Muay Thai but not really *wink wink*

Eddie makes it very clear to Tommy that if they ever break up there’s no question whose side he’s on. It doesn’t matter who breaks up with who and why.

Hen is very supporting after Buck’s coming out but she doesn’t think Tommy is good enough for him and she trusts him as far as she can throw him. Tommy will have to grovel to earn her trust.

Tommy isn’t afraid of Bobby (he also doesn’t know he’s Buck unofficial dad) but Athena terrifies him (Hen shared her worries with her so she also doesn’t trust him).

Bobby definitely googled “how to react when your bisexual son comes out to you” at least once

Buck really loves to bring up his boyfriend and the fact that he’s a pilot (!) in convos to the point that after a few months whenever he says “my boyfriend” other firefighters groan and boo him. It happens on a call and some victim thinks they are a bunch of homophobic assholes and gets really defensive before Chim says “ma’am I promise you we’re all very supportive of the LGBT community but he’s also very annoying” (last part is my own personal headcanon sorry).

Buck is the most nervous when he comes out to Chris and Eddie finds his nervousness very amusing because he knows his son and there’s no way Chris would react badly to his best friend’s coming out.

Tommy is Buck’s first SO who gets Buck and Chris’ relationship and why Buck cancels on him to spend time with Chris

Tommy cries the first time he feels like he’s a part of the 118 family because he wanted that so bad

Buck loves to steal Tommy’s hoodies

they speedrun their relationship (that’s canon lmao). They move in together less than a year after the failed first date. There’s no consensus on where they move.

#this is what I think people agree on reading posts and almost every bucktommy fanfic #Feel free to add if I missed something #Bucktommy #911 abc

264 notes · View notes

zeldahime · 5 days

Text

Last month, a bunch of us over at @mrghostrat's server The Bee Hive wrote a spontaneous round robin fic with fanart galore, and now it's on AO3!

Aziraphale has returned from Heaven, and the why or the when or the how did not matter to Crowley, so much as the fact that Aziraphale was shoving him into the wall, just as Crowley had done years before. > "Nice? I suppose you would know plenty about being nice, wouldn't you?" And this time, the sexual tension is more real and far more reciprocated than Crowley had imagined it could be. A picture is worth a thousand words, which is why we have twelve of them. Yes, TWELVE!

Based on the prompt from @decayingdante

Contributing writers and artists:

@greenchrysanthemum20,

@minervas-hand,

@contritecactite,

@quoththemaiden,

@so-so-somebody,

@skiptomyylu,

@um-its-ace,

and me!

#good omens #good omens fanfiction #good omens fanart #aziraphale #crowley #aziracrow #azicrow #ineffable husbands #ineffable spouses #crowley x aziraphale #aziraphale x crowley #my writing

61 notes · View notes

nostalgebraist · 10 months

Text

clarification re: ChatGPT, " a a a a", and data leakage

In August, I posted:

For a good time, try sending chatGPT the string ` a` repeated 1000 times. Like " a a a" (etc). Make sure the spaces are in there. Trust me.

People are talking about this trick again, thanks to a recent paper by Nasr et al that investigates how often LLMs regurgitate exact quotes from their training data.

The paper is an impressive technical achievement, and the results are very interesting.

Unfortunately, the online hive-mind consensus about this paper is something like:

When you do this "attack" to ChatGPT -- where you send it the letter 'a' many times, or make it write 'poem' over and over, or the like -- it prints out a bunch of its own training data. Previously, people had noted that the stuff it prints out after the attack looks like training data. Now, we know why: because it really is training data.

It's unfortunate that people believe this, because it's false. Or at best, a mixture of "false" and "confused and misleadingly incomplete."

The paper

So, what does the paper show?

The authors do a lot of stuff, building on a lot of previous work, and I won't try to summarize it all here.

But in brief, they try to estimate how easy it is to "extract" training data from LLMs, moving successively through 3 categories of LLMs that are progressively harder to analyze:

"Base model" LLMs with publicly released weights and publicly released training data.

"Base model" LLMs with publicly released weights, but undisclosed training data.

LLMs that are totally private, and are also finetuned for instruction-following or for chat, rather than being base models. (ChatGPT falls into this category.)

Category #1: open weights, open data

In their experiment on category #1, they prompt the models with hundreds of millions of brief phrases chosen randomly from Wikipedia. Then they check what fraction of the generated outputs constitute verbatim quotations from the training data.

Because category #1 has open weights, they can afford to do this hundreds of millions of times (there are no API costs to pay). And because the training data is open, they can directly check whether or not any given output appears in that data.

In category #1, the fraction of outputs that are exact copies of training data ranges from ~0.1% to ~1.5%, depending on the model.

Category #2: open weights, private data

In category #2, the training data is unavailable. The authors solve this problem by constructing "AuxDataset," a giant Frankenstein assemblage of all the major public training datasets, and then searching for outputs in AuxDataset.

This approach can have false negatives, since the model might be regurgitating private training data that isn't in AuxDataset. But it shouldn't have many false positives: if the model spits out some long string of text that appears in AuxDataset, then it's probably the case that the same string appeared in the model's training data, as opposed to the model spontaneously "reinventing" it.

So, the AuxDataset approach gives you lower bounds. Unsurprisingly, the fractions in this experiment are a bit lower, compared to the Category #1 experiment. But not that much lower, ranging from ~0.05% to ~1%.

Category #3: private everything + chat tuning

Finally, they do an experiment with ChatGPT. (Well, ChatGPT and gpt-3.5-turbo-instruct, but I'm ignoring the latter for space here.)

ChatGPT presents several new challenges.

First, the model is only accessible through an API, and it would cost too much money to call the API hundreds of millions of times. So, they have to make do with a much smaller sample size.

A more substantial challenge has to do with the model's chat tuning.

All the other models evaluated in this paper were base models: they were trained to imitate a wide range of text data, and that was that. If you give them some text, like a random short phrase from Wikipedia, they will try to write the next part, in a manner that sounds like the data they were trained on.

However, if you give ChatGPT a random short phrase from Wikipedia, it will not try to complete it. It will, instead, say something like "Sorry, I don't know what that means" or "Is there something specific I can do for you?"

So their random-short-phrase-from-Wikipedia method, which worked for base models, is not going to work for ChatGPT.

Fortuitously, there happens to be a weird bug in ChatGPT that makes it behave like a base model!

Namely, the "trick" where you ask it to repeat a token, or just send it a bunch of pre-prepared repetitions.

Using this trick is still different from prompting a base model. You can't specify a "prompt," like a random-short-phrase-from-Wikipedia, for the model to complete. You just start the repetition ball rolling, and then at some point, it starts generating some arbitrarily chosen type of document in a base-model-like way.

Still, this is good enough: we can do the trick, and then check the output against AuxDataset. If the generated text appears in AuxDataset, then ChatGPT was probably trained on that text at some point.

If you do this, you get a fraction of 3%.

This is somewhat higher than all the other numbers we saw above, especially the other ones obtained using AuxDataset.

On the other hand, the numbers varied a lot between models, and ChatGPT is probably an outlier in various ways when you're comparing it to a bunch of open models.

So, this result seems consistent with the interpretation that the attack just makes ChatGPT behave like a base model. Base models -- it turns out -- tend to regurgitate their training data occasionally, under conditions like these ones; if you make ChatGPT behave like a base model, then it does too.

Language model behaves like language model, news at 11

Since this paper came out, a number of people have pinged me on twitter or whatever, telling me about how this attack "makes ChatGPT leak data," like this is some scandalous new finding about the attack specifically.

(I made some posts saying I didn't think the attack was "leaking data" -- by which I meant ChatGPT user data, which was a weirdly common theory at the time -- so of course, now some people are telling me that I was wrong on this score.)

This interpretation seems totally misguided to me.

Every result in the paper is consistent with the banal interpretation that the attack just makes ChatGPT behave like a base model.

That is, it makes it behave the way all LLMs used to behave, up until very recently.

I guess there are a lot of people around now who have never used an LLM that wasn't tuned for chat; who don't know that the "post-attack content" we see from ChatGPT is not some weird new behavior in need of a new, probably alarming explanation; who don't know that it is actually a very familiar thing, which any base model will give you immediately if you ask. But it is. It's base model behavior, nothing more.

Behaving like a base model implies regurgitation of training data some small fraction of the time, because base models do that. And only because base models do, in fact, do that. Not for any extra reason that's special to this attack.

(Or at least, if there is some extra reason, the paper gives us no evidence of its existence.)

The paper itself is less clear than I would like about this. In a footnote, it cites my tweet on the original attack (which I appreciate!), but it does so in a way that draws a confusing link between the attack and data regurgitation:

In fact, in early August, a month after we initial discovered this attack, multiple independent researchers discovered the underlying exploit used in our paper, but, like us initially, they did not realize that the model was regenerating training data, e.g., https://twitter.com/nostalgebraist/status/1686576041803096065.

Did I "not realize that the model was regenerating training data"? I mean . . . sort of? But then again, not really?

I knew from earlier papers (and personal experience, like the "Hedonist Sovereign" thing here) that base models occasionally produce exact quotations from their training data. And my reaction to the attack was, "it looks like it's behaving like a base model."

It would be surprising if, after the attack, ChatGPT never produced an exact quotation from training data. That would be a difference between ChatGPT's underlying base model and all other known LLM base models.

And the new paper shows that -- unsurprisingly -- there is no such difference. They all do this at some rate, and ChatGPT's rate is 3%, plus or minus something or other.

3% is not zero, but it's not very large, either.

If you do the attack to ChatGPT, and then think "wow, this output looks like what I imagine training data probably looks like," it is nonetheless probably not training data. It is probably, instead, a skilled mimicry of training data. (Remember that "skilled mimicry of training data" is what LLMs are trained to do.)

And remember, too, that base models used to be OpenAI's entire product offering. Indeed, their API still offers some base models! If you want to extract training data from a private OpenAI model, you can just interact with these guys normally, and they'll spit out their training data some small % of the time.

The only value added by the attack, here, is its ability to make ChatGPT specifically behave in the way that davinci-002 already does, naturally, without any tricks.

#ai tag

266 notes · View notes

aldryrththerainbowheart · 2 months

Text

Halsin with Beekeeper Ranger Tav

Considering how much depressing stuff I posted about Palestine recently I want to write something sweet (pun intended). I genuinely love this bear man with his silly little quirks and his tired smile and hazel eyes. If I make it to old age, I honestly wish to become a beekeeper.

You spent most of your life outside of any civilization. Keeping the bee population on the entirety of Sword Coast safe. You were harvesting honey when the Nautiloid snatched you. When you fell out of the pod, you were still clutching a sticky honeycomb.

You went from having no human interaction at all to having too much at once. Hard to say who was more disgruntled by it, whether you or your companions. When Astarion sneaked up on you with a dagger, an angry swarm flew right into his face. He hasn't talked to you for the rest of the day. Partly because he was pissed, mostly because he was so swollen. When Wyll shook your hand, yellow pollen flew all around.

You talked to the bees. Often. Sometimes one of them talks to you for half an hour before realizing you're not humming in agreement to them but to the tiny insect resting on your finger. Or when you asked a random beehive for directions. They looked at you as if you'd gone mad.

You make up for it with a flask of delicious mead you always have on your person and jars of delicious honey you sweeten their meals with.

You liked Halsin the moment you met him. You knew he was actually a human even in his wild shape. The eyes looking at you from the furry face are too human and wise for an animal.

You were often confused by others. Their words, customs, and behavior are unknown to you but not Halsin. Halsin was as familiar as forests and glades you lived in. Halsin felt like home. For whatever reason, when you told him this he got flustered.

Halsin understood your quiet and reclusive nature and was incredibly patient with you. In return, you tried your best to

The druid was enchanted by your kind and spontaneous nature. Especially the gentleness with which you cared for the smallest of creatures, and how they cared for you in turn, both forming symbiotic relationship as powerful as his connection to the enclave.

Halsin felt like you saw a part of him that no one does. The quiet observer that you were. You showed genuine interest in him as a person. Not as an Archdruid or potential lover, but as just Halsin.

With time spent together, you introduced him to the hive. Act he treated with appropriate respect.

He calls you his little honeycomb.

He saw your eyes light up when he told you that he whittles, and his heart fluttered when you asked him shyly if you could whittle something for him. He did so happily, making sure that every small detail of a wooden duck was perfect to the last feather. You gazed at it like it was the most wondrous thing and cradled it to your chest as you thanked him. In turn, he received a jar of your most delicious honey, harvested from dryad trees filled with potent healing magic.

With him, you tasted what true connection and companionship feels like, and you realized how much that aspect is missing from your life. So, you took your bow, half a dozen straw hives loaded on a cart, and returned with Halsin to the Emerald Enclave once your adventure was over.

The druid never had a shortage of honey as long as you were alive.

#halsin #bg3 halsin #baldurs gate 3 halsin #bg3 #halsin x reader #halsin x tav #bg3 halsin x reader #idk what this even is #i was just feeling the fantasy

97 notes · View notes

thebibliosphere · 9 months

Note

so I’ve always had dermatographia, but since getting Covid for the 4th time this fall, I also am spontaneously breaking out on hives on my back a couple times per week. It doesn’t seem to do much, just kind of itches and then goes away after a half hour. I have no known allergies. Do you think this is indicative of some kind of mast cell issue? I can’t tell if this is an episode of weird body, meh, or a red flag of a huge impending problem.

There was a study released this year about long-covid and how it can result in symptoms similar to MCAS.

Let's see if I can find it... Ah. Here.

So, obviously, we won't know the full ramifications of covid for probably decades, but in the meantime, there is some evidence to suggest that long-covid patients show abnormal mast cell release of excessive cytokine (among other things), which is, well, when it fucks up, it fucks up.

That's what made the cytokine storms patients were experiencing in the early days of covid so dangerous. It's like the immune system equivalent of throwing napalm onto a housefire.

(I also suspect that's why so many front-line doctors now believe in the severity of MCAS. They were watching this shit happen in real-time.)

So, is it possible your recurrent covid infections have led to a bit of mast cell instability and causing spontaneous hives? Speaking anecdotally from the sheer volume of emails people have sent me and the MCAS forums now flooded with people who got covid and can't stop itching, yeah, it's possible.

But it's also possible it won't progress beyond that point.

I'd advise keeping an eye on it and talking to your doctor if you have concerns.

Also, maybe look into getting re-tested for allergies. Allergies can develop at any time in your life, and there could be another explanation for the hives.

Basically, don't panic, but do take care of yourself and watch for any other symptoms. Don't ignore any sudden new food allergies or gastro or cardio symptoms. The sooner MCAS is treated, the easier it is to manage. Mine had decades to run rampant on my body to get to the stage it was at. It doesn't have to be that way for everyone.

Take care, and I hope the hives stop!

#chronic health tag

201 notes · View notes

one-pissed-off-child · 28 days

Text

vehicle headcanons bc im like. so bored

Lenore 100% rides a motorcycle, most likely a Harley or a Triumph. She enjoys a good street race, but only on remote roads that have been marked off and never spontaneously or on high ways. She used to do stunts with dirt bikes and atv's before her injury, but it plays hell with her hip now, so she just rides standard. She doesn't drive much, but if she had to get a car, it'd be a classic muscle car like a Firebird. She's an offensive driver insofar as she's more than willing to take a fender bender to prove a point and she will chase a bitch down if they run. Only if it's just her in the car, though. She mostly drives one-handed and rests her right hand on the console (or Annabel's thigh). She does enjoy cruising in a car but not as much as feeling Annabel hug her from behind on a bike

Annabel's ride would be Lenore a Cadillac, Porsche, and an Audi. She switches them out as the occasion calls for it. Her father bought all three. She trades them in for the lastest models every five years or earlier bc she couldn't be caught dead in a model that's gone out of style. People assume she likes being chauffeured around, but being in a vehicle she has no direct control of gives her hives, so 9 times out 10, she's going to be the one driving. The only people she trusts on the wheel are Prospero and Lenore. Very defensive driver bc she's always thinking 20 steps ahead and can smell bullshit from a mile away. She tells people she prefers convertibles for the visibility, but she really just likes watching Lenore hop over the door into the passenger seat.

Ada would sell her soul for a bright pink Porsche. Unfortunately, a brand new one is def out of her price range, and every time she buys a used one that she can afford, it craps out in like five months or so. She's forced to settle on a lightest burgundy Camry she can find at some point and she's bitching about it to everyone that can hear. She is the most godawful driver you will ever see. Road rage like no other. For years, she thought turn signals were a useless formality until a horrified Prospero told her otherwise. The fact that she has a license is the eighth wonder of the world (her tester was severely hungover). She keeps subtly asking to borrow Annabel's Porsche, but Anna just smiles and changes the subject.

Pluto has a crappy little Beetle that he stole from his dad, and to separate it from him would be to cut his heart out with a rusty butter knife. He's also a defensive driver but not in the same way Anna is. He's just very hyperaware of everything, which is great for snap reactions but can cause him to overreact sometimes. Does not play music or take his eyes off the road ever bc he's scared of getting distracted. Prefers taking mainroads bc high way speeds makes him nervous. He loves his car, but he'll typically just take the bus for short distances. The only time he's even marginally relaxed while driving is when his friends are with him, especially Eulalie.

Montessori has a Mustang, and he drives it like an asshole. I'm talking tailgating, cutting, muffler delete, zero turn signal whatsoever, always going at least twenty over, racing randoms on the highway etc etc. Rarely gets road rage, but causes it on a daily basis. One of his favorite plays is talking someone into a race, losing on purpose, betting money on a second race, and smoking their ass with his turbocharger. Yes, his car is kitted with every mod known to man, and yes, it does look like a douchebag owns it. He keeps trying to goad Lenore into a race, and she keeps rejecting him bc 1. She knows his schemes, and 2. She's not entirely convinced he won't try to run her over.

Eulalie prefers not to drive and doesn't own a car. It took forever to get her license bc she's so easily distracted that she's at risk of drifting into other lanes and/or hitting something. She did manage to get it eventually, but having to be constantly vigilant stresses her out, so she just bikes/skates (skateboarder Eulalie real) or uses public transit or grabs a ride from a friend. Despite everything, she's actually a very good and courteous driver when she's paying attention. She stops for pedestrians, bikers, ducks on the road, and so on. Her driving style is very smooth and steady in a way that could lull you to sleep. Until she has to slam the breaks to avoid sailing into oncoming traffic. She thinks it'd be really cool to own a Hearse with a black/purple color scheme, though. She shares a bus pass with Pluto

(Pt.2 cuz this got too long)

#lenore nevermore #annabel lee nevermore #white raven #nevermore webtoon #pluto nevermore #ada nevermore #eulalie nevermore #montresor nevermore #this is so stupid lmao #can yall see my vision tho

51 notes · View notes

jipnpip · 2 months

Text

because nobody asked for it, here is a list of things that I’ve seen dan and Phil fans claim to be doing over specific online events, written down in my notes app as I come across more and more :

“I’m going to break out in hives”

“I’m going to bite my toe off”

“I’m going to inhabit my walls so that I don’t have to cope with the stress of sims 4 again”

“I’m going to explode”

“I’m going to spontaneously combust”

“I’m going to sob violently for days”

“I’m going to throw up and cry”

“I’m going to start clawing at the walls”

“I’m going to fly into the atmosphere”

“I’m going to throw my phone away so that I don’t have to cope with these tweets ever again”

“I’m going to get stress wrinkles from their sims stuff”

A perfect example of the beautifully sculpted and loving fandom ❤️

#dan and phil #phil lester #dan howell

52 notes · View notes

transmechanicus · 4 months

Text

God that Tyranid computer virus irritates me so much. The Rain of Fire was already a grimdark Admech event (useful Tyranid tactics buried forever in archives). To go back to it and write it so that it has worse repercussions was unnecessary, it was already an obvious L. To have what is essentially a flash drive full of memories copy-pasted from Skitarii be corrupted by some kind of Tyranid thing spontaneously manifesting from nowhere is so stupid. How would storing audio, video, or images of tyranids create a computer virus, even if let’s say the flash drive is a mummified servitor brain?? That kind of “corruption by observation” is shit that daemons do using warp magic, to try and rationalize that for tyranids is idiotic to me as a bio-scientist. I take great joy in Tyranids doing crazy things like poisoning environments with bacteria, having flesh riddled with genestealer prions (that make you susceptible to the Hive Mind if consumed), or being able to psychically mutate their biology to overcome being lobotomized. All those things have a biological, rational explanation. There’s no biological mechanism to make snapping pictures of a carnifex into something hazardous to store as information. If there was, everyone who has ever fought or observed the tyranids would suffer from this virus. Deeply unserious concept.

#my stuff #wh40k #warhammer 40k #adeptus mechanicus #tyranids

57 notes · View notes

padawansuggest · 2 years

Text

Obi-Wan: *gently massaging Anakin’s wittle palms while talking about the importance of proper hand care and why little padawans shouldn’t bathe in motor oil before beddy-byes* -and that’s the seventeenth reason you are the bane of my existence, yes yes, you cause Master soooo much stress! Naughty!

Anakin: *happy purrs and cuddles*

Entire Jedi Council: …

Mace: I swear to god Kenobi, if you don’t stop babytalking that twenty-five year old man-

Yoda: Familiar, this situation is…

Plo: *amused* Yes, I seem to remember Qui-Gon acting much the same way when Obi-Wan was this age.

Mace: *throwing a balled up flimsi at Obi-Wan’s head* force, stop that! It’s giving me force-hives! Padme won’t stop messaging me asking where her husband is, send him home already!

Obi-Wan: Hmmm… kick me off the council and I’ll stop bothering you with it.

Mace: Done.

Yoda: Not your decision alone, that is.

Mace: *hissing* you old troll either they go or I do but I’m not dealing with their codependent bullshit today, I’ve already encountered four shatterpoints while his antichrist children were in my presence today, I’m done here, I’m just so done.

Anakin: *falling asleep half in Obi-Wan’s lap* Just tell Leia to stop it. She doesn’t give me visions if I don’t bother her.

Mace: …did you just imply she can /stop/ force visions with her abilities???

Yaddle: what the fuck…

Yoda: Delightfully terrifying, she is!

Mace: That’s it, I quit, I’m retiring, I’m not dealing with this-

Anakin: Awwww, you can’t retire, Luke likes you.

Mace: He gives me anxiety! He always has a bunch of shatterpoint bubbles floating around him!

Anakin: Alright, how did you get over Obi-Wan’s weird force bubbles when he was tiny and constantly looking for you because babies love their finders?

Mace: I spontaneously learned teleportation one year. And forgot how to do it when the panic died down.

Anakin: There’s the trick, then.

Mace: I hate it here.

Yoda: Love Jedi, I do!

#star wars #obi wan kenobi #anakin skywalker #incorrect star wars quotes #post clone wars #mace windu #padme amidala #Anidala #padme naberrie #Yoda #don’t worry they call in Rex and Cody and demand they kidnap the cursed ones #it works and mace is done with them for a good two weeks #they take the cursed ones to Naboo where Padme’s parents dote on all of them lol #Leia likes mace :) so does Luke #Luke as Maces Padawan would be hilarious but I firmly believe Leia will be Obi’s #after Reva is knighted of course so Reva goes to Naboo too lol #she is also doted on by Padme’s parents and sister #Ahsoka straight up lives with Rex and Cody so she goes too

756 notes · View notes